Skip to content

Platform / Open Protocol

285 endpoints. Any AI agent.

Bring Claude, Copilot, your own. If it speaks MCP, it can use Spaceflow. The actions in your ERP become tools any model can call.

Book a demoAPI reference

MCP-native · REST and webhook fallbacks · OAuth2 · mTLS

agent.ts

typescript

1import { Client } from "@modelcontextprotocol/sdk";
2
3const sf = new Client({ name: "agent", version: "1.0" });
4await sf.connect({ transport: "sse", url: "mcp.spaceflow.tech" });
5
6const tools = await sf.listTools();
7// → 285 tools, scoped to the calling user
8
9await sf.callTool("procurement.quote.compare", {
10  rfqId: "RFQ-2847",
11});

currently running against this stack

6 models · 0 lock-in

Claude Opus 4.7

Anthropic

GPT-5

OpenAI

Gemini 2.5 Pro

Google

Llama 4 405B

Meta

Mistral Large 3

Mistral

In-house · BERT-x

self-host

What this replaces

Closed integrations are an expiring asset.

The real cost of being locked into one model or one vendor shows up eighteen months from now, when the better model is somewhere else and the integration has to be rebuilt.

Every AI vendor wants to own the stack

Lock-in

The pattern is familiar. The integration ships fast because it only has to work with one model. A year later the better model is somewhere else, and the rebuild starts. Open protocols turn that switch into a config change.

Custom connectors age fast

Tech debt

A new endpoint here, a deprecated field there. Bespoke integrations are most fragile where they meet the LLM, because the LLM half changes every quarter. MCP gives the integration a stable contract on the model side.

Switching models means rebuilding the loop

6 months / rebuild

If your prompts, your retry logic, and your tool definitions are coupled to one provider, every migration is a project. Standardize the interface and the model becomes a hot-swap.

What 'open' actually means here.

  1. MCP

    Model Context Protocol native

    Spaceflow speaks the protocol the major AI labs are converging on. Tools, resources, prompts, sampling, all of it. No proprietary shims.

  2. REST

    REST + webhook fallback

    If you can't speak MCP yet, the same 285 actions are available as REST endpoints with idempotency keys and webhooks for async responses.

  3. OAUTH2

    Identity passthrough

    Every call carries the calling user's identity downstream. Your existing access rules in SAP, Oracle, and Dynamics do the gatekeeping.

  4. GATEWAY

    Single audited choke point

    All traffic flows through our gateway. Every call is policy-checked, schema-validated, and written to an append-only log.

Works with

Already running against.

ClaudeGPT-5Gemini 2.5CopilotLlama 4Mistral LargeCustom in-house models